https://www.w3.org/wiki/LargeTripleStores
And in production, Wikidata has scaled to 13B+ triples so far using Blazegraph that is public to the world: https://scholia.toolforge.org/statistics Sadly, http://sparqles.ai.wu.ac.at/ is down but it had a list of smaller endpoints you could check for examples of what dataset sizes you can host publicly with a good uptime. You can excavate some of its data on https://web.archive.org/web/*/http://sparqles.ai.wu.ac.at//* (use text/html MIME filter). I think http://lod-a-lot.lod.labs.vu.nl/ shut down its SPARQL endpoint in favour of an TPF interface<https://linkeddatafragments.org/data/> but it had 20B+ triples AFAIK queryable over SPARQL at some point of time. /Andrew